frasercrmck wrote: > > which in its header uses this strange **asm** method of calling LLVM > > intrinsics directly. > > That's something that's always surprised me it works. It's rather unsafe (you > can bypass immarg validation for instance). Plus asm callsites get infected > with overly conservative attributes (like convergent, which you can't remove)
Yeah, good point. > > It should maybe just do: OpenCL builtin -> CLC builtin -> clang builtin? > > That's the simplest way to go I've updated the patch to do just that, using the builtins. I'll update the description accordingly. https://github.com/llvm/llvm-project/pull/119596 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
